Skip to contents

Audit R code for risky external references (deterministic pipeline).

Source: R/code_audit.R
dot-audit_code_impl.Rd

Runs the full first-pass audit: AST extraction -> tool-layer path whitelist -> deterministic read of whitelisted source files -> (optionally) feed the read text to the Data Shield reviewer. The reviewer NEVER gets a read/write/shell tool: this function decides what to read (code, not LLM), bounds it to whitelisted in-project source files, and only hands the reviewer vetted text.

Usage

.audit_code_impl(
  code,
  shield = NULL,
  project_root = getwd(),
  max_bytes = 100000L,
  max_files = 20L
)

Arguments

code

Character. The R code to audit (untrusted; parsed, not run).

shield

Optional DataShield whose reviewer reviews the referenced file contents. NULL -> report references only (no content review).

project_root

Directory the whitelist confines reads to.

max_bytes

Per-file read cap (avoid feeding huge files to the reviewer).

Value

A list: static_paths, dynamic, allowed (paths read), blocked (list of path+reason not read), reviews (per-file reviewer verdicts when a shield is given), risk (overall: "none"/"review"/"block").